In this paper, a novel quantization scheme for the wavelet coefficient
s is introduced. Using the wavelet packet transform (WPT) and lattice
vector quantization (LVQ), we present here a new lattice optimization
scheme based on an accurate model for the distribution of the wavelet
coefficients. The model is based on the generalized gaussian distribut
ion (GGD). A least squares algorithm on a non-linear function of the s
hape parameter is formulated to estimate the model parameters. The pro
posed algorithm adapts to non-stationarity in input images and to give
n bit rates. Compared to other wavelet-based algorithms, the technique
proposed here results in higher reconstructed image qualities for ide
ntical bit rates. (C) 1997 Published by Elsevier Science B.V.